fix(openclaw): configure native Bedrock embeddings for memory search - #75
Conversation
Add memorySearch config pointing to the amazon-bedrock plugin's native embedding support (amazon.titan-embed-text-v2:0). Uses the same IAM role credential chain — no external proxy needed. Tested locally: openclaw memory index --force succeeds, memory search returns results with Bedrock Titan embeddings. Fixes the 'memory search is paused — index needs rebuild' error on fresh installs.
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: f00062e1a7
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| "models": {"providers": {"amazon-bedrock": {"baseUrl": f"https://bedrock-runtime.{bedrock_region}.amazonaws.com", "auth": "aws-sdk", "api": "bedrock-converse-stream", "models": bedrock_models}}}, | ||
| "plugins": {"entries": {"amazon-bedrock": {"enabled": True, "config": {"discovery": {"enabled": True}}}}}, | ||
| "agents": {"defaults": {"model": {"primary": PRIMARY_MODEL, "fallbacks": [FALLBACK_MODEL]}, "workspace": f"{home}/.openclaw/workspace", "compaction": {"mode": "safeguard"}, "heartbeat": {"model": HEARTBEAT_MODEL, "target": "telegram", "every": "30m", "lightContext": True, "isolatedSession": True}, "maxConcurrent": 4, "subagents": {"maxConcurrent": 8}}}, | ||
| "agents": {"defaults": {"model": {"primary": PRIMARY_MODEL, "fallbacks": [FALLBACK_MODEL]}, "workspace": f"{home}/.openclaw/workspace", "compaction": {"mode": "safeguard"}, "memorySearch": {"provider": "bedrock", "model": "amazon.titan-embed-text-v2:0"}, "heartbeat": {"model": HEARTBEAT_MODEL, "target": "telegram", "every": "30m", "lightContext": True, "isolatedSession": True}, "maxConcurrent": 4, "subagents": {"maxConcurrent": 8}}}, |
There was a problem hiding this comment.
Because this switches memory search to the native Bedrock embedding provider, the OpenClaw pack no longer needs the proxy, but fresh deploys still install registry dependencies first: packs/registry.yaml still lists bedrockify for openclaw, and deploy/bootstrap.sh runs those deps before the pack installer. In any fresh OpenClaw install where the bedrockify download or daemon health check fails, bootstrap aborts before this config is ever written even though the new path does not use bedrockify. Please remove or update that dependency alongside this config change.
Useful? React with 👍 / 👎.
1 participant
Adds
memorySearchconfig to use the amazon-bedrock plugin's native embedding support (amazon.titan-embed-text-v2:0). Same IAM credentials, no proxy needed.Tested locally:
No bedrockify proxy required — the
@openclaw/amazon-bedrock-providerplugin already implements thememoryEmbeddingProviderscontract natively.